I recently reviewed a clinic’s plan to deploy AI decision support. It hit every word on the AI bingo card: LLM this, agent that, RAG the other… A critical piece was missing: they had no concept of data infrastructure.
The clinic operates with siloed systems — clinic management software, an EHR storing lab data as PDFs, plus various intake forms. They’re attempting to climb the mountain of AI Enlightenment from the Cave of Siloed Data, not realising the raw material isn’t there.
Healthcare is fundamentally a logistics problem; the doctor should be an air traffic controller. Instead, they’re organising suitcases and counting vegan meals. This is the AI mirage. We need to liberate data before processing it.
Amateurs talk strategy; professionals talk logistics.
The reality? We’re in 1995. Most labs send PDFs — some structured, most not — with randomly changing formats that challenge OCR extraction. API connectivity barely exists due to warped incentives.
Looking ahead, two key developments
Agentics: Small, capable specialist automation handling discrete workflow tasks, often collaborating with other agents.
Citizen developers: Non-technical domain experts driving transformation with user-friendly AI tools when IT can’t keep pace.
The AI mirage persists when we view AI as a universal remedy rather than a tool requiring robust data foundations. We must shift from hype to practicality — only then can we bridge the gap between AI’s promise and its application in healthcare.
